Testing the primary and convergent retrieval model of recall: Recall practice produces faster recall success but also faster recall failure

Abstract

The primary and convergent retrieval (PCR) model assumes that the act of successful recall not only boosts associations between the item and retrieval cues but additionally strengthens associations within the item (i.e., between the features of an item), speeding the rate of information retrieval from memory. The latter effect is termed intra-item learning and is a unique benefit of recall practice (i.e., the "testing effect"). Prior work confirmed the prediction that recall practice produces faster subsequent recall than restudy practice even if accuracy is higher following restudy. The current study replicated this result, but also examined the downside of recall practice: that after a failure to recall during practice, participants will be faster in their failure to recall on a subsequent recall test. This prediction was confirmed in a multisession cued recall experiment that collected accuracy and recall latency measurements for no practice, recall practice, or restudy, with an immediate or delayed final test. The linear ballistic accumulator model was fit to latency distributions, and model comparison determined that these effects reflect differences in drift rates, as predicted by the PCR model.

Keywords: Cognitive modeling; Cued recall; Episodic memory; Retrieval practice.

Figure 1:

Illustration of different directed associations that are learned when studying (or restudying) the word pair ‘table – bank’ (solid arrows) or when successfully recalling ‘bank’ in response to the cue word ‘table’ (dashed arrows). During initial study, a participant might create a mental image of a picnic table by a river bank to episodically conjoin the words. This produces directed associations from these words to the mental image. Restudy strengthens these forward associations, but this does not necessarily enhance recall of ‘bank’ upon retrieval of this mental image, which relies backward associations. In contrast, successful cued recall practice involves activation of the mental image in response to ‘table’ and then activation of ‘bank’ in response to this mental image. According to the PCR memory model learning rule, this establishes directed associations from this mental image to the word ‘bank’ as well as directed associations from ‘table’ to ‘bank’, with both of these pathways boosting subsequent cued recall performance (both accuracy and latency) in response to the cue word ‘table’.

Figure 2:

Illustration of the convergent retrieval process and intra-item learning. Nodes represent episodic item features; filled nodes are active features, and unfilled nodes are inactive features. Arrowed paths represent associative connections. Solid arrows represent associative connections that activate subsequent features in the convergent retrieval process. The activation rule in this illustration requires two input connection from active nodes to cause an inactive node to become active. Dashed arrows represent new intra-item associative connections learned from retrieval. Left Column: Successful recall on a practice test produces intra-item learning, enabling faster retrieval on the final test. Right Column: Unsuccessful recall on a practice test also entails intra-item learning, but in this case intra-item learning results in faster failure to recall on the final test.

Figure 3:

Panel A: Example of the “Recall” vs “Can’t Recall” decision made on each test trial of the current experiment. Panel B: Schematic representation of a decision between the “Recall” and “Can’t Recall” alternatives as described by the LBA model. The accumulator that intersects the response threshold first is the chosen alternative, and the response time is the amount of time elapsed before the response threshold is met. In this example, the “Recall” accumulator reaches the threshold first, and is the response given on this simulated trial.

Figure 4:

Outline of the experimental design, with the temporal structure of the sessions flowing left to right, and then top to bottom.

Figure 5:

Performance across conditions. Larger, darker points represent averages across participants. Smaller grey points represent observations from individual participants. Error bars represent +/− one standard error of the mean, estimated using the subject-normalized method of Morey (2008). Top Row: Recall accuracy on the final cued recall test. Bottom Row: Average decision latency on the final cued recall test. Incorrect latencies reflect trials where participants indicated they could not recall the target item. Correct latencies reflect trials where participants could recall the target item, and subsequently provided the correct word as a response.

Figure 6:

LBA Model RT quantiles, together with empirical quantiles estimated directly from the observed data. The quantile values were estimated at the .1, .3, .5, .7, and .9 quantiles of the RT distributions. No quantile functions for correct “Recall” responses are presented for the delayed final test incorrect practice test condition because there were an insufficient number of responses for this situation.

Figure 7:

Average drift rates across participants for each condition from the "$v$ Free" (bottom row) and "$v$ and $b$ Free" models (top row). Error bars represent +/− one standard error of the mean within each condition.